train ResNet-50 Alternating training Erroe

[chenxu93]

Hi， I want to traing ResNet-50 using Alternating training but meet a error I do not know how solve it,could somebody help?thank you!

Cannot copy param 0 weights from layer 'rpn_conv/3x3'; shape mismatch. Source param shape is 512 1024 3 3 (4718592); target param shape is 256 1024 3 3 (2359296). To learn this layer's parameters from scratch rather than copying from a saved net, rename the layer.

by the way ,I use the pretrained model resnet50_rfcn_final.caffemodel,and I changed the realtive parameters in the pt files. I can train ResNet-50 end2end_ohem by change the layer rfcn_cls to my_rfcn_cls,but how can I do to solve this problem?

[chenxu93]

ok I hace solved this problem，i just use the pretrain model provide in https://onedrive.live.com/?authkey=%21AAFW2-FVoxeVRck&id=4006CBB8476FF777%2117887&cid=4006CBB8476FF777.

[Small93]

Cannot copy param 0 weights from layer 'rfcn_cls'; shape mismatch. Source param shape is 1029 1024 1 1 (1053696); target param shape is 147 1024 1 1 (150528). To learn this layer's parameters from scratch rather than copying from a saved net, rename the layer. *** Check failure stack trace: *** ./experiments/scripts/rfcn_end2end.sh: 行 57: 4209 已放弃 (核心已转储) ./tools/train_net.py --gpu ${GPU_ID} --solver models/${PT_DIR}/${NET}/rfcn_end2end/solver.prototxt --weights data/imagenet_models/${NET}-model.caffemodel --imdb ${TRAIN_IMDB} --iters ${ITERS} --cfg experiments/cfgs/rfcn_end2end.yml ${EXTRA_ARGS} i train ResNet-50 end2end_ohem. can you help me?

[Small93]

i rename the layer 'rfcn_cls'. it works. thanks.